Top Bananas | Tag Archives: machine learning

Understanding the Threat of Artificial Intelligence

Posted on April 18, 2018 by admin Posted in Programming Leave a comment

What it is: Elon Musk, CEO of Tesla and SpaceX, constantly warns people about the dangers of artificial intelligence.

Many people see artificial intelligence as nothing more than positive progress. That’s why you see so much research occurring in the self-driving car market but artificial intelligence has long been around in other forms as well. Every time you buy a product on Amazon, Amazon’s algorithms analyze your purchase and the purchases of people buying similar products. Then Amazon can offer you other products you might be interested in. This not only increases sales, but also makes Amazon’s algorithm smarter for the next user.

Banks and credit card companies have been using artificial intelligence for years to track down credit card fraud. Email accounts use spam filters to keep you from getting overwhelmed by junk email. Casinos use facial recognition algorithms to look for professional gambrels who aren’t allowed to gamble any more. Artificial intelligence has been infiltrating every aspect of our lives to the point where getting rid of it is not practical.

The huge problem with artificial intelligence is that when we allow programs to adapt and learn on their own, humans no longer fully understand how the program works any more. As artificial intelligence grows in sophistication, it’s likely to start controlling our lives in unintended ways, which likely won’t be for the betterment of the human race. Instead of deliberately attacking people like fictional artificial intelligence computers and robots might do, real artificial intelligence can simply allow humanity to wither away through sheer neglect. Either way means it’s bad result for people.

Think of the bias already inherent in police reporting. In poor neighborhoods, police report more crimes because they record those crimes. In rich neighborhoods, police are less likely to report crimes perpetrated by rich people. So the end result is that the data skewers towards more crime in poor areas and less crime in rich areas, which can convince a program to pour more resources into poor areas to combat crime through increasingly draconian measures.

Computers simply do what they’re programmed to do, but if they can start modifying their own behavior, nobody knows how they’ll behave any more. The more we give up control to artificial intelligence, the less control we’ll have over our own destiny. That doesn’t necessarily mean the extinction of the human race, but it does mean we’ll be dependent on something we no longer control or understand.

Artificial intelligence will only get better over time. Whether it can ever get smart enough to equal human intelligence is another question, but if it does, then it will likely surpass human intelligence as well. If that happens, what will people do with computers that are smarter than they are and that nobody knows how to control?

Questions like thees are the reason why Elon Musk is concerned about the future of artificial intelligence because without controls put in place ahead of time, or even thinking about safety, artificial intelligence will likely charge ahead without any safety measures put in place at all. Thinking safety last is a prescription for disaster.

To learn more about the dangers of artificial intelligence and why we should be more concerned than we are, read “Our Final Invention,” a book Elon Musk recommends everyone read. When you realize computer scientists are charging ahead building artificial intelligence without thinking about safety issues first, you realize that everyone’s safety will be in jeopardy in the long run.

<br />

Privacy on a Chip and Differential Privacy

Posted on December 27, 2017 by admin Posted in Hardware, Software Leave a comment

What it is: Differential privacy is a way to mine data without exposing the privacy of the people generating that data.

In the world of artificial intelligence and machine learning, d at a is king. The more data you have, the more the computer can detect patterns and get “smarter.” That’s why the focus of most technology companies lies with collecting as much data as possible, which is why Google collects your personal browsing habits to make a better search engine. Because Google has so much data everyday, their search engine can get progressively more accurate over time, making it nearly impossible for competitors like Microsoft’s Bing to ever catch up.

However, the huge drawback of most machine learning projects involves sending data to the cloud so a company like Google can use it. This cause two problems. First, such data collection requires an Internet connection. Lose that Internet connection and the device suddenly loses its artificial intelligence. Second, such data collection also reveals the activities of individuals. Essentially this means you’re willingly giving up your privacy to help companies like Google sharpen their machine learning algorithms.

To avoid exposing users’ privacy, Apple uses something called differential privacy. The essential details is that differential privacy seeks to maximize data collection and analysis while minimizing the ability to pinpoint where that data came from. In other words, differential privacy protects your privacy.

In addition to protecting your privacy, Apple is developing its own neural engine on a chip. So instead of relying on a data center to provide artificial intelligence, Apple’s products will use a neural engine to provide the artificial intelligence whether your device has an Internet connection or not.

Apple’s approach to artificial intelligence is to protect your privacy while also making their devices smarter without relying on an always-on Internet connection. Whether this will prove better than current machine learning practices that send data to the cloud remains to be seen. As long as Apple’s solution is good enough while also protecting user privacy, it may be preferable than a “smarter” artificial intelligence device that exposes your activities so anyone (think the government) can examine your activities and use it against you.

To compete, rivals like Google and Microsoft will need to develop their own artificial intelligence on a chip. Then again, they may simply keep doing what they’re doing and assuming people will willingly trade their privacy for results. Some people don’t mind giving up their privacy, but if you do, then you may be wary of other approaches to machine learning that does not use differential privacy.

Optimized Apps vs. Cross-Platform

Posted on December 24, 2017 by admin Posted in Software Leave a comment

What it is: Pixelmator Pro offers graphics editing capabilities at a fraction fo the cost of Adobe Photoshop.

In every software category, a leader emerges. For years, that leader maintains its grip on the market until something changes. Then that leader starts fading away and a new leader emerges.

In the operating system world, CP/M-80 was the business standard because it ran WordStar and VisiCalc, two early dominant programs in their market. CP/M-80 eventually faded when CP/M-86 failed to catch on because MS-DOS was cheaper and came free with most IBM PC compatible computers. WordStar faded when WordPerfect took over the word processing market, and VisiCalc faded when Lotus 1-2-3 appeared.

What’s notable about all three dominant leaders is that they didn’t lose their market because a better competitor came along, but a different competitor appeared. CP/M-80 had the Z80 processor market locked up but when IBM PCs standardized on Intel processors like the 8088, MS-DOS had a chance to take the market away from CP/M-86. Suddenly CP/M-80 wasn’t a factor any more and MS-DOS could compete against CP/M-86 on equal terms. Both operating systems were essentially the same, but MS-DOS came free with IBM PC compatibles while CP/M-86 did not. That alone spelled the difference and led to massive success for MS-DOS.

WordStar wasn’t a bad word processor, but WordPerfect came along and offered toll-free support, which shifted the balance to WordPerfect and led people to abandon WordStar. VisiCalc worked just fine until Lotus 1-2-3 included graphics and ran optimized on MS-DOS while VisiCalc was making the transition from CP/M-80 to CP/M-86 and MS-DOS. given the fact that Lotus 1-2-3 offered more features and was optimized for MS-DOS (unlike VisiCalc), people soon abandoned VisiCalc and used Lotus 1-2-3 instead.

In today’s graphics world, Adobe Photoshop reigns supreme. If you’re a professional graphics artist, you have to use Photoshop. That’s why Adobe can afford to charge a subscription to all Photoshop users so they get a steady stream of revenue all year long. If you don’t want to stay locked into Photoshop’s subscription, you now have a new choice on the Macintosh called Pixelmator Pro.

Pixelmator Pro probably won’t replace Photoshop, but it can steal away a sizable share of the market. First of all, Pixelmator Pro is completely optimized for the Macintosh, using Apple’s graphics framework called Metal to speed up displaying and manipulating images. Photoshop is not optimized for the Macintosh because it’s cross-platform to run on both Windows and the Macintosh. That means less optimized code for both Windows and the Macintosh.

Second, Pixelmator Pro offers machine learning to take advantage of Apple’s new CoreML framework that lets you add machine learning to your programs. This is a feature that’s easy to implement in Macintosh software but not so easy to implement on Windows software. That means Adobe can’t easily add similar machine learning features to Photoshop.

Given Pixelmator Pro’s lower price and its optimization for the Macintosh, it only makes sense to use Pixelmator Pro unless you absolutely need to use Photoshop. It’s likely professional graphics artists will use both Photoshop and Pixelmator Pro at the same time because of Pixelmator Pro’s low cost. Then they can choose which program to use based on what they need to do.

Pixelmator Pro may never displace Photoshop because it doesn’t run on Windows, so Photoshop will likely remain a standard. However, Pixelmator Pro will gain a huge Macintosh following regardless. As Pixelmator Pro evolves and the optimization features make it faster and easier to use than Photoshop, expect more graphics artists to rely on Pixelmator Pro more often. As Windows fades from dominance, expect Photoshop’s influence to fade as well.

Pixelmator Pro may never take over Photoshop’s lead, but it will make a mark of its own, and part of that mark may be taking over Photoshop’s lead in the Macintosh market because Pixelmator Pro is optimized for the Macintosh and Photoshop is not.

Pixelmator Pro Offers Machine Learning Tools

Posted on December 1, 2017 by admin Posted in Software Leave a comment

What it is: Pixelmator Pro recently became available and uses machine learning to help identify items in an image.

For years, graphic artists relied on Adobe Photoshop to edit images. The problem as that Photoshop kept getting more expensive. Then Adobe stopped selling copies of Photoshop altogether and went to a subscription model that required people to pay a monthly fee. Many graphic artists simply paid that monthly fee but many others rebelled at the idea of constantly paying to use Photoshop. That’s why many Macintosh users simply switched from Photoshop to rivals such as Affinity Photo and Pixelmator.

Pixelmator is relatively inexpensive (under $30) but is over 10 years old, which means it was likely written in Objective-C. The Pixelmator programmers decided to rewrite their app completely using Swift, which gave them chance to redesign their program to create a much more powerful graphics program dubbed Pixelmator Pro ($59). Despite its higher price over the original Pixelmator, it’s still far cheaper than subscribing to Adobe Photoshop. Best of all, Pixelmator uses the latest Apple technologies, such as Metal, to provide optimized graphics performance that Adobe Photoshop cannot match because Photoshop’s code must be cross-platform to support Windows and Macintosh users.

So not only is Pixelmator Pro optimized for the Macintosh, it’s also cheaper than Photoshop. One example of Pixelmator Pro taking advantage of Apple’s latest technology is the inclusion of machine learning to identify objects in an image. As more macOS programs get written entirely in Swift, expect macOS programs to offer superior performance compared to cross-platform rivals that can’t offer similar levels of performance on the Macintosh.

As any programmer can tell you, code gets bloated over time and Pixelmator’s original 10-year old design means lots of code likely isn’t optimized for today’s Macintosh. That’s why rewriting Pixelmator Pro entirely in Swift to take advantage of Apple’s frameworks like Metal will make a huge difference in Pixelmator Pro’s performance. As the first version, there’s likely to be flaws in Pixelmator Pro but you can expect the company to quickly address these issues and keep improving the program over time.

If you’re a graphic artist tired of paying for Adobe Photoshop, Pixelmator Pro may be a viable option. Given its low cost and optimization for the Macintosh, you may find you don’t need Adobe Photoshop after all.

Using Machine Learning

Posted on September 5, 2017 by admin Posted in Software Leave a comment

What it is: Pixelmator announced a new professional graphics editor called Pixelmator Pro that uses machine learning.

For the longest time, Adobe Photoshop ruled the graphics editing world. The problem of Photoshop was that it was expensive, it was cross-platform for Windows and macOS so it wasn’t really optimized for either operating system, and Adobe switched to a subscription model so you had to buy a software subscription to use Photoshop. For many graphics professionals, Photoshop is the only alternative.

However, Pixelmator is a low-cost alternative graphics editor that does much of what most people need. Now the company that brought Pixelmator has introduced a new professional version called Pixelmator Pro. Although the cost hasn’t been revealed yet, one unique feature is that Pixelmator Pro uses Apple’s latest machine learning features. Presumably this machine learning feature will allow Pixelmator Pro to help people edit images faster and more accurately than before, but this could be the first major use of machine learning in a major application so it will be interesting to see how well it works.

The real key to machine learning isn’t just to use trained models but to allow users to train and update the machine learning model through use of the program. While this likely isn’t possible yet, it’s what will make machine learning far more valuable to adapt to the user rather than offer a canned response every time.

Pixelmator Pro offers real competition to Adobe Photoshop, and machine learning just makes this program even more intriguing. Best of all, Pixelmator Pro is optimized solely for macOS so it should run better and faster than Adobe Photoshop. The only question is whether Pixelmator Pro can offer all the features professional graphics users need.

Whatever the case, the graphics editing world is getting full of rivals and alternatives to Photoshop. Beyond Pixelmator Pro, there’s Affinity Photo along with Adobe’s own Photoshop Elements, a simpler version of Photoshop for the average person.

Photoshop may be the dominate graphics editor today and maybe even tomorrow, but now you won’t have to rely on Photoshop any more if you don’t want to. Pixelmator Pro will be coming this fall and this may give you a chance to leave Photoshop once and for all.

To learn more about Pixelmator Pro, click here.

Apple AI Researchers Win Prestigious Award

Posted on August 26, 2017 by admin Posted in Programming, Software Leave a comment

What it is: Artificial intelligence (AI) researchers at Apple recently won a Best Paper Award at the 2017 Conference on Computer Vision & Pattern Recognition, one of the most sought after prizes in the field.

For the longest time, Apple refused to let their researchers publish papers. The rationale behind this was to keep the competition from following what Apple was doing, but the drawback to this was that many AI researchers refused to work at Apple because they didn’t want to keep their research private. Many researchers also wondered whether secrecy could actually hurt Apple by keeping their researchers from sharing ideas with others in the AI field.

Eventually, Apple let their researchers publish papers and in the first year of submitting papers, Apple researchers won a Best Paper Award for a computer vision paper. What’s notable about this award is that it proves that Apple isn’t behind in AI research as many critics have suggested.

Google has been testing self-driving cars for years, long before Apple started so people naturally thought Google has an insurmountable lead. While starting early can be a benefit, such an early head start can be lost if researchers from other companies come up with more creative and innovative solutions. AI research isn’t a matter of starting first and assuming you’ll always stay ahead forever. AI research is a new field that everyone is studying so there’s no fixed way to insure a lead in any part of AI research.

Spending more money and hiring more researchers won’t always be the answer because smaller teams of researchers with less money can still find an innovative solution that larger, better funded teams may overlook. AI research is a brand new field that’s ripe for pioneers from everywhere. The fact that Google got started with self-driving cars earlier than most everyone else is an advantage, but not an insurmountable one. Chances are good that researchers all over the world are ahead of Google in certain areas of autonomous technology while Google is ahead of everyone else in other areas.

Apple was first in commercially developing natural voice language processing with Siri, but despite being first, Amazon and Google seem to have caught up. Still, Siri has the advantage in recognizing more languages than its rivals, but its rivals appear better at understanding the context of spoken commands better.

So it’s likely that AI research will continue advancing from all different companies. Expect Apple to continue promoting AI. Already Apple is promoting machine learning for creating apps but this is just a first step. Right now, Apple’s machine learning option lets you take a trained machine learning file and use it. However, this trained machine learning file seems fixed. Once you use it, it never learns again. In the future, expect Apple to allow your apps to create their own machine learning models and constantly modify and improve its features.

AI is coming to smartphone and tablet apps along with computer programs for Windows and macOS. AI is the next rapidly growing field of the computer industry so expect to see more companies throwing around the terms of artificial intelligence on their products. AI is going to be a crucial element of software and computer development in the future. Apple’s latest prize award simply shows that Apple is not behind its rivals in AI research after all.